graph neural networks (GNNs) have been extensively developed for graph
representation learning in various application domains. However, similar to all
other neural networks models, GNNs suffer from the black-box problem as people
cannot understand the mechanism underlying them. To solv